AgentSkillsCN

implement-plan

将头脑中的想法、分享的内容或既定的计划转化为技术方案,并辅以严格验证。当您已经准备好一份完整的实施计划,可以按部就班地执行;在多阶段实施过程中,以严谨的验证手段确保每一步都符合预期;按照成功标准跟踪进度;在复杂功能的开发中,采用分阶段执行的方式;或者在特定的文件与行号基础上实施功能,并配套完善的测试流程时,可使用此技能。

SKILL.md
--- frontmatter
name: implement-plan
description: Implement technical plans from thoughts/shared/plans with verification. Use when you have a completed implementation plan ready to execute systematically, executing multi-phase implementations with methodical verification, tracking progress against success criteria, working on complex features with phase-by-phase execution, or implementing features with specific file:line references and testing procedures.

Implement Technical Plan

Execute implementation plans phase by phase with automated verification at each milestone.

Related Skills

  • validate-plan: Run after completing implementation to verify all success criteria
  • create-handoff: If pausing mid-implementation, document current state and next steps
  • iterate-plan: If implementation reveals the plan needs updates

Workflow

1. Load the Plan

  • Ask for path if not provided: thoughts/shared/plans/YYYY-MM-DD-*.md
  • Read complete plan, extract phases, success criteria, dependencies
  • Verify prerequisites: git status, git log -1, pre-flight checks
  • Note any assumptions or prerequisites mentioned in the plan

2. Execute Each Phase

For each phase:

Implement:

  • Make changes according to plan specification
  • Reference specific file:line numbers from plan
  • Follow existing code patterns
  • Commit incrementally with descriptive messages

Example commit message:

code
Phase 1: Add configuration validation layer

- Implement ConfigValidator class (config/validator.ts)
- Add unit tests for validation rules
- Ref: thoughts/shared/plans/2025-12-23-config-refactor.md

Verify automatically:

bash
[tests command - e.g., npm test, pytest, go test]
[type check command - e.g., tsc --noEmit, mypy]
[lint command - e.g., eslint, ruff, golangci-lint]
[build command - e.g., npm run build, make]

Report results:

  • Which checks passed (summary)
  • Which checks failed (full error output)
  • Any unexpected failures

Manual verification:

  • Follow testing steps from plan
  • Test each scenario in success criteria
  • Verify edge cases and error handling

3. Phase Gate

After success criteria met:

code
✓ Phase [N]: [Phase Name]

Implemented:
- [specific changes made]
- [files modified]

Verification Results:
- Automated checks: All passing
- Manual verification: Complete

Commits: [hash(es)]

Proceed to Phase [N+1]? (waiting for confirmation)

Wait for explicit user confirmation before proceeding.

4. Final Verification

After all phases:

  1. Run full validation suite - All automated checks one more time
  2. End-to-end testing - Verify complete feature works
  3. Document completion:
code
✅ Implementation Complete: [Feature Name]

Plan: thoughts/shared/plans/[filename]

Phases Executed:
- Phase 1: [Name] ✓ [commit hash]
- Phase 2: [Name] ✓ [commit hash]

Success Criteria Verified:
- Automated: All passing
- Manual: All verified

Files modified: [count]
Commits: [count] | Branch: [name]

Next Steps:
- [Follow-up items]

Handling Issues

Automated checks fail:

  1. Show complete error output
  2. Analyze what went wrong
  3. Propose solutions or ask for guidance
  4. Ask: "Fix this, troubleshoot further, or try different approach?"

Manual testing reveals problems:

  1. Document the issue clearly
  2. Determine if in scope for current phase
  3. Fix immediately or note as follow-up

Plan unclear:

  1. Quote the ambiguous text
  2. Propose interpretation
  3. Wait for confirmation before proceeding
  4. Consider using iterate-plan if significant ambiguity

Key Principles

  • One phase at a time - Complete all criteria before advancing
  • Reference plan continuously - Don't drift; flag deviations
  • Verify automatically first - Then manual testing
  • Explicit confirmation between phases - Never auto-advance
  • Document issues immediately - Don't let problems accumulate